Diff from mypy_primer, showing the effect of this PR on open source code:

prefect (https://github.com/PrefectHQ/prefect)
+ src/prefect/locking/memory.py:194: note:     def get(self, Never, None, /) -> None
+ src/prefect/locking/memory.py:205: note:     def get(self, Never, None, /) -> None
+ src/prefect/records/memory.py:176: note:     def get(self, Never, None, /) -> None

steam.py (https://github.com/Gobot1234/steam.py)
+ steam/ext/commands/utils.py:43: note:          @overload
+ steam/ext/commands/utils.py:43: note:          def get(self, str, None, /) -> _VT | None

streamlit (https://github.com/streamlit/streamlit)
+ lib/streamlit/runtime/state/session_state.py: note: In member "get_id_from_key" of class "KeyIdMapper":
+ lib/streamlit/runtime/state/session_state.py:314:9: error: Returning Any from function declared to return "str"  [no-any-return]

operator (https://github.com/canonical/operator)
- ops/_private/harness.py:1193: error: Argument 2 to "get" of "dict" has incompatible type "None"; expected "Mapping[str, str]"  [arg-type]
+ ops/_private/harness.py:1193: error: Incompatible return value type (got "dict[str, str] | None", expected "Mapping[str, str]")  [return-value]

beartype (https://github.com/beartype/beartype)
+ beartype/_check/error/errcause.py:551: error: Incompatible types in assignment (expression has type "Callable[[ViolationCause], ViolationCause] | None", variable has type "Callable[[ViolationCause], ViolationCause]")  [assignment]
+ beartype/_check/error/errcause.py:552: error: Unused "type: ignore" comment  [unused-ignore]

mypy (https://github.com/python/mypy)
+ mypy/partially_defined.py:510: error: Never apply isinstance() to unexpanded types; use mypy.types.get_proper_type() first  [misc]
+ mypy/partially_defined.py:510: note: If you pass on the original type after the check, always use its unexpanded version

prefect and stream.py are no actual change. operator and beartype are both changing an existing error from an argument error to a return type error.

I dug into the streamlit error. It's something hinky. This produces a new error with this change:

from typing import Any

def get_id_from_key(key_id_mapping: dict[str, str], default: Any) -> str:
    return key_id_mapping.get("key", default)

But this version:

from typing import Any

def get_id_from_key(key_id_mapping: dict[str, str], default: Any) -> str:
    result = key_id_mapping.get("key", default)
    return result

Which assigns the return value to a variable before returning, causes mypy to generate Returning Any from function declared to return "str" both with and without this change.

Okay, what's going on there is that with the call in the return position, mypy uses the context of the return type being str while checking the overloads. overload 1 is discard as non-matching. Overload two is (str, _VT) -> _VT, which is (str, str) -> str. The current third overload is (str, _T) -> _T | _VT which the context infers to (str, str) -> str. That's the same for both main and this MR.

The new overload in this MR get (str, None) -> str | None. Mypy then runs a check to see if there's any ambiguity, and finds none for main, but does find ambiguity for this version.

The difference between here and assigning to result first is that in that case, mypy doesn't have the context to infer str for _T. We can give it that context with result: str = key_id_mapping.get("key", default) in which case there's no error for either main or this MR, not because the ambiguity went away but because there's no error for "Assigned Any to a typed variable" the way there is for "returning Any from function".

So that's all probably fine and sensible.

Short version of the mypy diff is that self.type_map.get(o.expr, None) evaluates to builtins.object before being passed to isinstance_proper_hook, while with this MR it evaluates to Union[mypy.types.Type, None] and the plugin considers mypy.types.Type to be improper.

I think this is similar to the streamlit error in that mypy uses context. For streamlit, it matched Any to the context of str normally, and here it's matching 'None' to the context of 'object' and skipping the union. We don't reach that with this MR applied - we match None to None first and don't reach the final branch of the overload. In a normal context, that's fine, and we'll match the union of Type and None down to object a little later, but the mypy plugin is checking before that happens in this case.